Back to EveryPatent.com
United States Patent |
5,117,401
|
Feintuch
|
May 26, 1992
|
Active adaptive noise canceller without training mode
Abstract
An active adaptive noise canceller that inserts delays in the weight update
logic of an adaptive filter employed by the canceller to make the filter
stable. It has been found that there is a great deal of flexibility
regarding the selection of the delay values. This insensitivity permits
designing the delays in advance, and not having to adjust them to
different situations as they change, thus no longer requiring a training
mode. The canceller dramatically reduces the amount of hardward needed to
perform active adaptive noise cancelling, and eliminates the need for the
training mode, which in some applications, including automobiles, for
example, can be as objectionable as the noise sources that are to be
suppressed.
Inventors:
|
Feintuch; Paul L. (Covina, CA)
|
Assignee:
|
Hughes Aircraft Company (Los Angeles, CA)
|
Appl. No.:
|
568289 |
Filed:
|
August 16, 1990 |
Current U.S. Class: |
367/135; 367/1; 367/137; 367/901; 381/71.11 |
Intern'l Class: |
H04B 015/00 |
Field of Search: |
367/901,137,1,135
381/71,94
364/574,724.19
181/206
|
References Cited
U.S. Patent Documents
4677676 | Jun., 1987 | Erikasson | 381/71.
|
4677677 | Jun., 1987 | Eriksson | 381/71.
|
4736431 | Apr., 1988 | Allie et al. | 381/71.
|
Primary Examiner: Pihulic; Daniel T.
Attorney, Agent or Firm: Denson-Low; Wanda K.
Claims
What is claimed is:
1. An active adaptive canceller for use in suppressing noise signals
derived from a noise source, said active adaptive canceller comprising:
a noise sensor;
an acoustic sensor;
an acoustic output device;
delay means coupled to the noise sensor for delaying the noise signals
generated thereby by a preselected time delay; and
adaptive filter means having a plurality of inputs coupled to the noise
sensor, the acoustic sensor, and the delay means, and an output coupled to
the acoustic output device;
wherein the delay means causes the active adaptive canceller to be stable
and to not require a training mode.
2. The active adaptive canceller of claim 1 wherein the adaptive filter
means comprises a plurality of adjustable filter weight inputs, and weight
update logic circuitry coupled to the plurality of adjustable filter
weight inputs and to the delay means for receiving output signals from the
acoustic sensor and delayed output signals from the delay means and for
adjusting the filter weights applied to the adjustable filter weight
inputs.
3. The active adaptive canceller of claim 1 wherein the adaptive filter
means and delay means comprises:
first adaptive filter means having an input and an output;
second adaptive filter means having an input and an output;
adder means coupled to the outputs of the first and second adaptive filter
means for combining the output signals provided thereby to provide
filtered output signals and for applying the filtered output signals to
the output device;
first delay means coupled to the first adaptive filter means for delaying
the filtered output signals coupled thereto by a first predetermined time
delay; and
second delay means coupled to the second adaptive filter means for delaying
the noise signals coupled thereto by a second predetermined time delay.
4. The active adaptive canceller of claim 3 wherein the first and second
predetermined time delays are substantially the same.
5. The active adaptive canceller of claim 1 wherein the adaptive filter
means and delay means comprises:
first adaptive filter means having an input and an output and including a
plurality of adjustable filter weight inputs;
second adaptive filter means having an input and an output and including a
plurality of adjustable filter weight inputs;
adder means coupled to the outputs of the first and second adaptive filter
means for combining the output signals provided thereby to provide
filtered output signals and for applying the filtered output signals to
the output device;
first weight update logic circuitry coupled to the first adaptive filter
means for receiving input signals comprising the filtered output signals
and output signals from the acoustic sensor and for adjusting the filter
weights applied to the adjustable filter weight inputs of the first
adaptive filter means;
second weight update logic circuitry coupled to the second adaptive filter
means for receiving input signals comprising the background noise signals
and output signals from the acoustic sensor and for adjusting the filter
weights applied to the adjustable filter weight inputs of the second
adaptive filter means;
first delay means coupled to the first weight update logic circuitry for
delaying the filtered output signals coupled to the first weight update
logic circuitry by a predetermined time delay; and
second delay means coupled to the second weight update logic circuitry for
delaying the background noise signals coupled to the second weight update
logic circuitry by a predetermined time delay.
6. An active adaptive canceller for use in suppressing noise signals
derived from a noise source, said active adaptive canceller comprising:
a noise sensor adapted to sense the noise signals;
an acoustic sensor;
an acoustic output device;
an adaptive filter coupled between the noise sensor and the acoustic output
device;
delay means coupled to the noise sensor for delaying the noise signals
generated thereby by a preselected time delay; and
weight update logic circuitry coupled between the the adaptive filter means
and the delay means for receiving output signals from the acoustic sensor
and delayed output signals from the delay means and for adjusting the
filter weights applied to the adjustable filter weight inputs of the
adaptive filter;
wherein the delay means causes the active adaptive canceller to be stable
and to not require a training mode.
7. An adaptive canceller for use in eliminating noise from a system
comprising a noise sensor, a speaker and an microphone that function in
the presence of background noise signals, said adaptive canceller
comprising:
a first adaptive filter having an input and an output and including a
plurality of adjustable filter weight inputs;
a second adaptive filter having an input and an output and including a
plurality of adjustable filter weight inputs;
means for combining the output signals provided by the first and second
adaptive filters to provide filtered output signals and for applying the
filtered output signals to the speaker;
first weight update logic circuitry coupled to the first adaptive filter
for receiving input signals comprising the filtered output signals and
output signals from the microphone and for adjusting the filter weights
applied to the adjustable filter weight inputs of the first adaptive
filter;
second weight update logic circuitry coupled to the second adaptive filter
for receiving input signals comprising the background noise signals and
output signals from the microphone and for adjusting the filter weights
applied to the adjustable filter weight inputs of the second adaptive
filter;
a first delay circuit coupled to the first weight update logic circuitry
for delaying the filtered output signals coupled to the first weight
update logic circuitry by a predetermined time delay; and
a second delay circuit coupled to the second weight update logic circuitry
for delaying the background noise signals coupled to the second weight
update logic circuitry by a predetermined time delay.
8. An adaptive canceller for use in eliminating noise from a system
comprising a noise sensor, a speaker and a microphone that function in the
presence of background noise signals, said adaptive canceller comprising:
first adaptive filter means having an input and an output and including a
plurality of adjustable filter weight inputs;
second adaptive filter means having an input and an output and including a
plurality of adjustable filter weight inputs;
adder means coupled to the outputs of the first and second adaptive filter
means for combining the output signals provided thereby to provide
filtered output signals and for applying the filtered output signals to
the speaker;
first weight update logic circuitry coupled to the first adaptive filter
means for receiving input signals comprising the filtered output signals
and output signals from the microphone and for adjusting the filter
weights applied to the adjustable filter weight inputs of the first
adaptive filter means;
second weight update logic circuitry coupled to the second adaptive filter
means for receiving input signals comprising the background noise signals
and output signals from the microphone and for adjusting the filter
weights applied to the adjustable filter weight inputs of the second
adaptive filter means;
first delay means coupled to the first weight update logic circuitry for
temporally delaying the filtered output signals coupled to the first
weight update logic circuitry by a predetermined fixed time delay; and
second delay means coupled to the second weight update logic circuitry for
temporally delaying the background noise signals coupled to the second
weight update logic circuitry by a predetermined fixed time delay.
9. An active adaptive canceller for use in suppressing acoustic noise
signals generated by an acoustic noise source, said active adaptive
canceller comprising:
a first sensor for receiving said noise signals and providing a signal
representative of said noise signals;
an output device for generating acoustic cancelling signals for cancelling
said acoustic noise signals;
a second sensor for receiving the acoustic noise signals and said acoustic
cancelling signals to provide combined acoustic signals;
a delay means for delaying the signals representative of said noise signals
by a preselected time delay to provide delayed signals representative of
said noise signals, said time related to the combined delay attributed to
the transfer function of the output device and the second sensor; and
an adaptive filter having its signal input coupled to receive said signals
representative of said noise signals, said adaptive filter having its
weight update logic coupled to receive the delayed signals representative
of said noise signals and the output of said second sensor, the output of
said adaptive filter coupled to said output device.
10. An active adaptive canceller as recited in claim 9 further including a
second delay means for delaying signals at its input by a preselected time
delay to provide delayed signals at its output, said time of the delay
related to the combined delay of the transfer function of the output
device and the second sensor; and wherein said adaptive filter comprises:
a first adaptive filter;
a second adaptive filter; and
an adder having at least two inputs and an output;
the second adaptive filter having its signal input coupled to receive said
signals representative of said noise signals, said adaptive filter having
its weight update logic coupled to receive the delayed signals
representative of said noise signals and the output of said second sensor
to receive, the output of said second adaptive filter coupled to one input
of said adder;
the output of said adder coupled to said output device and to the input of
said second delay means;
the first adaptive filter having its signal input coupled to receive the
output of said adder, said first adaptive filter having its weight update
logic coupled to receive the output of said second delay means and the
output of said second sensor.
Description
BACKGROUND
The present invention relates generally to adaptive noise cancellers, and
more particularly, to active adaptive noise cancellers that do not require
a training mode.
Current active adaptive noise cancellation systems use the so called
"filtered-X LMS" algorithm, and require that a potentially very
objectionable training mode be used to learn the transfer function of a
speaker and microphone employed in the systems.
All previously known active noise cancellers utilize the training mode to
learn the transfer functions of the speakers and microphones used in their
systems. As the physical situation changes, training must be redone. For
example, in an automobile application, the training mode needs to be
re-initiated every time a window is opened, or another passenger enters
the car, or when the vehicle heats up during the day.
By way of introduction, the objective in active noise cancellation is to
generate a waveform that inverts a nuisance noise source and suppresses it
at some point in space. This is termed active noise cancelling because
energy is added to the physical situation. In conventional noise
cancelling applications, such as echo cancelling, sidelobe cancelling, and
channel equalization, a measured reference is transformed to subtract out
from a primary waveform. In active noise cancelling, a waveform is
generated for subtraction, and the subtraction is performed acoustically
rather than electrically.
In the most basic active noise cancellation system, a noise source is
measured with a local sensor such as an accelerometer or microphone. The
noise propagates both acoustically and structurally to a point in space,
such as the location of the microphone, at which the objective is to
remove the components due to the noise source.
The measured noise waveform at its source is the input to an adaptive
filter, the output of which drives the speaker. The microphone measures
the sum of the actual noise source and speaker output that have propagated
to the point where the microphone is located. This serves as the error
waveform for updating the adaptive filter. The adaptive filter changes its
weights as it iterates in time to produce a speaker output that at the
microphone that looks as much as possible (in the minimum mean squared
error sense) like the inverse of the noise at that point is space. Thus,
in driving the error waveform to have minimum power, the adaptive filter
removes the noise by driving the speaker to invert it. Thus the term
active cancellation.
In conventional applications of adaptive cancellation, the input to the
adaptive filter is called the reference waveform. The filter output is
electrically subtracted from the desired waveform channel (called the
primary waveform) which is corrupted by the noise to be removed. The
difference (called the error) is directly observable and is fed back to
update the adaptive filter using a product of the error and the data into
the adaptive filter in an LMS weight update algorithm.
Although the error summation in an active cancellation system is performed
acoustically in the medium, it is possible to represent this system by an
equivalent electrical model. The adaptive filter output is passed through
the speaker transferer function and is then subtracted from the channel
output to form the error which is observable only through the microphone
transfer function. Thus the observable error is not directly based on the
adaptive filter output, but on the adaptive filter output passed through
the speaker transfer function. In addition, the error difference is not
directly observable, but is only observable through the microphone
transfer function. Therefore, there are two major structural differences
between the active noise cancelling problem and conventional adaptive
cancellation. Direct application of the LMS algorithm within this
configuration results in filter instability, which is clearly
unacceptable. For that reason, all active noise cancelling applications
utilize the "filtered-X" LMS algorithm instead, which requires a training
mode.
In the training mode the transfer function of the speaker-microphone
combination is estimated. A broadband noise source (different from the
noise sources described above) is input to both the speaker and a separate
adaptive filter that is different from the one used for adaptive
cancellation (this filter does not drive the filter and its output is not
used at all). The microphone output is then subtracted from the adaptive
filter output to form the error waveform which updates the filter. The
adaptive filter attempts to make its output look like the
speaker-microphone output, thus estimating the cascaded transfer
functions. The adaptive filter is updated with the straight LMS algorithm,
in that the adaptive filter output is directly subtracted from the
waveform it is trying to estimate (the output of the speaker-microphone),
and the error for updating the LMS algorithm is directly observable as
well. The converged adaptive filter in steadystate has a transfer function
denoted by G(SM), which will have been learned in the training mode. The
filter G(SM) is then used in the filtered-X configuration to compensate
for the speaker and microphone effects.
An adaptive filter employing the filtered-X LMS algorithm uses two adaptive
filters, one of which is slaved to the other. The first adaptive filter is
used only to form the weights that are used in the slaved filter. The
output of the first adaptive filter is not used. The first adaptive filter
has its input filtered by the estimated speaker-microphone transfer
function, G(SM), which was learned during the training mode. Thus the
slave adaptive filter update is based on the filtered data, rather than
the data itself, and the error, which is not the direct subtraction of the
filter output from the waveform channel output. Since the filter input
(reference waveform) is often called the X-channel in adaptive filter
literature, this configuration is called the "Filtered-X LMS" algorithm.
This algorithm is discussed in the book entitled "Adaptive Signal
Processing," by B. Widrow et al, Prentice-Hall, 1985.
In addition, if the microphone appears in both the waveform channel and
speaker portions of the circuit prior to error subtraction, if the speaker
or microphone contain zeros (which they very likely will), or if the
waveform channel or microphone contain poles (which is also very likely),
then the adaptive filter will have to produce poles to either undo the
speaker-microphone zeros or to transform the noise to model the waveform
channel-microphone poles. The limitation here is in the basic
finite-impulse-response (FIR) structure of the LMS adaptive filter, which
produces only zeros. The LMS adaptive filter can approximate a pole by
having a large number of weights, but this results in slow convergence (a
severe limitation in practical applications) and is expensive. Thus the
need exists to modify the LMS algorithm configuration to adjust its
weights based on something other than the error-data product since that is
not available, and to produce poles, or remove the need to produce poles.
If in the filtered-X LMS algorithm, G(SM) is made part of the noise source
measurement, G(SM).sup.-1 is needed on the slave adaptive filter input so
as not to change the situation from that of the just-described filter. The
speaker-microphone transfer function, which was estimated to be G(SM) in
the training mode, is undone by the equivalent of G(SM).sup.-1 in front of
the slaved adaptive filter. The zeros of the speaker-microphone will be
exactly cancelled by the poles of G(SM).sup.-1. This eliminates one of the
reasons the adaptive filter needs to produce poles. It does nothing about
the poles in either the waveform channel or the microphone. More
importantly, it provides the adaptive algorithm with the correlated inputs
it needs to converge. The adaptive filter on the actual input data is then
slaved to have the weights formed using the filtered-X.
A logical question at this stage is whether an adaptive filter that can
produce poles implicitly within its structure would be more appropriate
for this problem. A recursive adaptive filter, which has a feed-forward
and feed-backward adaptive section produces both poles and zeros. It may
be used instead of the adaptive filter first discussed above. The problem
is that the recursive adaptive filter needs to be updated by the error,
which is the direct difference between the adaptive filter output and the
waveform channel output. This is not the case with the active canceller,
where the error is only observable through the speaker-microphone. In
addition the waveform channel output is modified by the inverse of the
speaker transfer function. Thus G(SM).sup.-1 is needed to provide the
recursive LMS algorithm with the error waveform it requires to properly
update the feed-forward and the feed-backward weights. It has been found
in simulations, that if G(SM).sup.-1 is not inserted, the recursive LMS
filter is also unstable. Thus, although the recursive LMS algorithm allows
the adaptive filter to produce the required poles, it still requires a
training mode to fully implement the algorithm.
Therefore, the primary objective of the invention is to eliminate the need
for the training mode, in active adaptive cancellation systems, for both
those that can and cannot produce poles. It is also an objective to
develop an alternative to estimating the speaker-microphone transfer
function and having to invert it in an adaptive canceller. There are
several practical motivations for this, aside from the complexity of the
system. The training mode is very awkward in many situations. For example,
in an automobile noise quieting problem, the car occupants are not going
to appreciate an irritating loud white noise in the interest of quieting
future noise. In addition, the training mode would need to be re-initiated
every time the situation in the vehicle changed in a way that could alter
the speaker-microphone transfer function, such as opening a window, adding
another passenger, the car heating up in the sun, and so forth. What is
needed is an alternative to the training mode that provides the system
with the correlations that are needed for the LMS or the recursive
adaptive filter algorithm to converge while operating over a wide range of
variations in the parameters associated with that alternative.
Consequently, there is a need for a new active adaptive canceller system
that does not require training, and therefore has much more practical
utility.
SUMMARY OF THE INVENTION
In accordance with the principles of the present invention, the present
active adaptive noise canceller provides for the use of either LMS or
recursive adaptive filters in "conventional" adaptive filter
configurations. There is no need for training modes to estimate
speaker-microphone transfer functions, or for the use of additional
filters as slaved filters required in the "filter-X" LMS configuration,
which is used to keep the adaptive filter stable. The filter is made
stable instead by the insertion of a delay value in the logic that
performs the calculation for the update of the adaptive filter weights.
The delay value approximates the delay in the combined speaker-microphone
transfer function, without requiring estimation of the entire
speaker-microphone transfer function. It has been found that there is a
large range of flexibility regarding the selection of the delay value, all
of which maintain stability of the adaptive canceller. This insensitivity
permits designing the delays in advance to cover the full range of
expected variations in almost any application, and not having to adjust
them to different situations as they change. As a result, the present
noise canceller no longer requires the training mode, which in many
applications for human comfort can be as objectionable as the noise
sources that the system is installed to suppress. In addition, the present
invention dramatically reduces the amount of hardware needed to perform
active adaptive noise cancelling, by no longer needing the " filtered-X"
configuration with its extra slaved adaptive filters to ensure filter
stability.
BRIEF DESCRIPTION OF THE DRAWINGS
The various features and advantages of the present invention may be more
readily understood with reference to the following detailed description
taken in conjunction with the accompanying drawings, wherein like
reference numerals designate like structural elements, and in which:
FIG. 1 shows a basic prior art adaptive noise canceller configuration;
FIG. 2 shows a generalized active adaptive noise canceller in accordance
with the principles of the present invention that does not require a
training mode;
FIG. 3 shows the "unwrapped" phase response of the system of FIG. 2 with no
delay and with a 13 sample delay; and
FIG. 4 shows a recursive active adaptive noise canceller in accordance with
the principles of the present invention that does not require a training
mode employing delays in the weight update logic; and
FIGS. 5-9 show results of simulations performed on the canceller of the
present invention.
DETAILED DESCRIPTION
With reference to FIG. 1, it shows a prior art active noise cancellation
system 10. In this basic active noise cancellation system 10, a noise
source 11 is measured with a local noise sensor 17 such as an
accelerometer or microphone. The noise propagates both acoustically and
structurally to a point in space, through what is termed a channel 15,
such as the location of the microphone 12, at which the objective is to
remove the components due to the noise source 11.
The measured noise waveform at its source is the input to an adaptive
filter 13, the output of which drives a speaker 14. The microphone 12
measures the outputs that propagate to the point where the microphone 12
is located. This serves as the error waveform for updating the adaptive
filter 13. The adaptive filter 13 changes its weights as it iterates in
time to produce a speaker output at the microphone 12 that looks as much
as possible (in the minimum mean squared error sense) like the inverse of
the noise at that point in space. Thus, in driving the error waveform to
have minimum power, the system 10 removes the noise at the microphone 12
by driving the speaker 14 to invert it.
In order to overcome the limitations of conventional noise canceller
systems such as those using the last mentioned principles, FIG. 2 shows a
generalized active adaptive noise canceller 20 in accordance with the
principles of the present invention that does not require a training mode.
The active adaptive noise canceller 20 comprises a sensor, such as a
microphone 12, that senses outputs of the speaker 14 and the channel 15.
Output signals from the microphone 12 are coupled to weight update logic
22 which is a portion of the adaptive filter 13. Noise from the noise
source 11 is sensed by the sensor 17 and coupled as an input to the
adaptive filter 13 and to a delay means 21, whose output is coupled to the
weight update logic 22. The output of the weight update logic 22 is
adaptive to drive the adaptive filter 13 whose output is coupled to the
speaker 15. The output of the speaker 14 and channel 15 are summed in an
adder 23 as shown in the electrical equivalent circuit of FIG. 2, but are
really combined acoustically by the microphone 12 in actual operation of
the canceller 20. The use of the delay means 21 renders the system 20 of
FIG. 2 stable. Simulations that will be discussed below indicate that a
wide range of delay values may be employed in the delay means 21 while
keeping the canceller 20 stable.
The principle exploited in the present invention is that the instability of
the conventional adaptive canceller for applications of active noise
cancellation, is due to its inability to compensate for the phase shifts
due to the speaker 14 and microphone 12 transfer functions. The canceller
20 is stable if the weight update logic 22 for the adaptive filter 13
includes the delay means 21 on the data portion of the weight update
calculation. A large range of values of this delay, encompassing the full
range expected in practice for any particular application, provides a
stable canceller 20, so that it need not be trained as in the filtered-X
canceller. This property holds for either a finite-impulse-response (FIR)
filter as used in LMS adaptive cancellers, or for the
infinite-impulse-response (IIR) or recursive adaptive filter cancellers,
as will be discussed in more detail below.
Results of simulations are presented herein that demonstrate the behavior
of the canceller 20 present invention. The simulations show that adaptive
filters are unstable without the delays, and are stable with the inclusion
of the delay means 21 in the adaptive filter 13 in accordance with the
principles of the present invention. In addition the simulations show that
one need not know the exact delay value to ensure stability, but that a
large range of values suffice. This robust character with respect to the
critical element of the present invention is what enables the removal of
the training mode.
The condition for stability requires that the phase of the product of the
speaker-microphone transfer function fall inside the regions between
2n.pi.-.pi./2 and 2n.pi.+.pi./2 for n=0, .+-.1, .+-.2, and so on. The
simulations show that the insertion of the delay 21 on the data portion of
the weight update extends the portions of the spectrum over which this
stability condition is met. If the input is bandpass filtered to the
portion of the band over which cancellation is desired, then the addition
of the delay 21 permits stability over that band by significantly
expanding the stability region. Without the delay 21, the canceller 20 is
not stable. The simulations show this behavior, for both finite impulse
response (FIR) LMS configurations of the canceller 20, and for infinite
impulse response (IIR) or recursive implementations of the canceller 20.
It is important to note that if the adaptive filter 13 needs to produce
poles, then the LMS algorithm can only approximate the pole by having a
large number of filter taps. The recursive filter can actually make poles
in its response, and can therefore provide a better steady state solution,
i.e. more cancellation, with fewer taps. However, an important aspect of
the present invention is not whether poles are needed in the final
transfer function of the adaptive filter 13, but that the filter 13 must
be stable in order to converge to its steady state solution, whether it
needs poles or not. The present invention allows use of FIR or IIR
adaptive filters 13 in simple canceller configurations by making them
stable via the insertion of the delays in the weight updates.
FIG. 3 is a graph that illustrates the stability region of the canceller 20
of FIG. 2, having phase in pi radians along the ordinate and frequency in
Hertz along the abscissa. FIG. 3 shows the "unwrapped" phase response of
the canceller 20 of FIG. 2 with no delay and with a 13 sample delay. FIG.
3 is also illustrative of the properties of various filter configurations
in which the principles of the present invention may be employed. These
will be discussed in more detail below.
A computer model was developed to investigate the active noise cancellation
system shown in FIG. 2. The purpose of the model was to demonstrate
canceller stability. For simplicity, the signal processing computations of
the model were implemented in the digital discrete-time domain. Since the
transfer functions of the speaker 14 and microphone 12 are critical in
determining stability, special care was taken to preserve the frequency
response characteristics of these analog functions when mapped into their
discrete-time equivalences.
A speaker transfer function was selected. The amplitude and phase response
functions of the speaker are such that the speaker frequency response is
limited to the approximate band of 50 to 3000 Hz. This is a reasonable
model of a typical inexpensive small speaker. In a similar manner, a
simple sixth order bandpass Butterworth filter was used to model the
microphone 12.
The next step was to determine the values of the delay to be inserted for
stability. The combined phases of the speaker 14 and microphone 12 (with
many 2.pi. discontinuities) must be "unwrapped" to yield a continuous
function of frequency. The solid line in FIG. 3 shows the effect of the
unwrapping on the phase characteristic of the speaker-microphone
combination with no delay. The stability condition requires the unwrapped
phase of the speaker-microphone transfer function to fall inside
(2n.pi.-.pi./2, 2n.pi.+.pi./2), n=0, .+-.1, .+-.2, . . . , which are the
stippled regions in FIG. 3. The dashed curve in FIG. 3 is the unwrapped
phase with a delay value of 13 samples. The solid curve in FIG. 3 displays
stability regions from approximately DC to 4.25 Hz, from 25 to 45 Hz, and
from 100 to 170 Hz.
A bulk delay has a phase response that is a straight line with slope
proportional to the delay. Thus, there is a limited range of frequencies
for which the bulk delay can stabilize the composite phase response of the
canceller 20. Therefore, there are phase characteristics where the
stability condition can never be achieved with just the insertion of bulk
delay. For the example shown in FIG. 3, no delay value yields algorithm
stability in the band 40 to 70 Hz. On the other hand, with delays,
stability is extended to the frequency region far above 170 Hz.
It was also investigated whether the range of delay values for which the
recursive LMS adaptive noise canceller 20 is effective is sufficiently
large to encompass physical changes that one would expect in a typical
application. If the range is sufficiently large, then one delay value in
the middle of this range may be selected, and the need for the training
mode is removed. The following simulation results show a remarkable
flexibility in the selection of the delay value. It was found that for an
input signal containing a tone as well as broadband noise, with the tone
at -3 dB, in that it contains half the input power, the canceller response
drops to -25 dB in less than 0.1 second.
The significant feature of the canceller 20 and simulation examples
presented herein is that in no case was a training mode employed. The
delay means 21 was employed to update the weights of the adaptive filter
13. In addition, the delay value may be varied over as many as four time
samples without changing the basic performance of the system 20, which
provides good, stable cancellation.
It can be concluded that the present invention, using recursive adaptive
filters that produce poles and zeros, may be used to provide rapid, stable
and significant cancellation without a training mode if the delay means 21
are inserted in the data channels that are used to form the weight updates
for the adaptive filter 13.
With reference to FIG. 4, it shows an electrical equivalent circuit of a
noise cancellation system 30 that includes a recursive LMS adaptive
canceller 40 in accordance with the principles of the present invention.
The system 30 comprises the channel 15 (typically air) that is the
transmission path for noise, and the speaker 14. Adder 16 represents the
summation of the acoustic output of the speaker 14 and the noise
transmitted by way of the noise propagation channel 15. The combined
signal (shown as the output of the adder 16) is sensed by the microphone
12. The output of the microphone 12 provides inputs to the recursive LMS
adaptive canceller 40 of the present invention.
The canceller 40 includes first and second LMS adaptive filters 41, 42
whose respective outputs are coupled to inputs of an adder 43, whose
output is coupled to the input of the speaker 14, and which comprises the
output of the canceller 40. The error feedback inputs to the canceller 40
provided by the microphone 12 are coupled to first and second weight
update logic circuits 44, 45, and the outputs of the first and second
weight update logic circuits 44, 45 provide weight values for the first
and second adaptive filters 41, 42, respectively. The input to the speaker
14 (that is, the output of adder 43) is also coupled as an input to the
first adaptive filter 41 and is coupled through a first delay 46 to the
first weight update logic circuit 44. The signal from the noise source 11
is coupled through the sensor 17 as a signal input to the second adaptive
filter 42 and is coupled a second delay 47 to the second weight update
logic circuit. as an input to the second adaptive filter 42, and is
coupled through a second delay 47 to the second weight update logic
circuit 45.
The recursive LMS adaptive noise canceller 40 of the present invention adds
the delays 46, 47 in the data path of a conventional recursive LMS filter.
The delays 46, 47 provide inputs to the weight update logic circuits 44,
45 that compute the adaptive filter weights. The delay values that are
chosen approximately compensate for the delay that the speaker-microphone
transfer function places on the error path. The innovation provided by the
present invention is the use of the delays 46, 47 to delay the inputs to
the weight update logic circuits 45, 46. In the recursive adaptive
canceller 40 in FIG. 3, the updates to the feed-forward and feed-backward
weights use delayed data sequences, rather than undelayed values. The use
of undelayed values as updates to the feed-forward and feed-back weight is
described in the article entitled "An Adaptive Recursive LMS Filter," by
P. L. Feintuch, IEEE Proceedings, Vol. 64, No. 11, November 1976. Without
the use of the delays 46, 47, the active cancellation system 30 is
unstable. With delays that are near the values of the delays caused by the
speaker 14 and microphone 12, the system 30 is stable. The recursive LMS
adaptive noise canceller 40 then corrects for spectral transformations
that are needed.
With regard to the above-mentioned simulations, presented below are results
of simulations for specific canceller types incorporating the principles
of the present invention. These canceller types include infinite impulse
response (IIR) recursive adaptive filters and the finite impulse response
(FIR) LMS adaptive filters.
Using the LMS adaptive filter structure shown in FIG. 2, the filter is
unstable with a delay value of zero, but is stable for 6 units of delay in
both the feed-forward and feed-backward weight updates. FIG. 5 shows a
power versus frequency graph for the case of any input to the canceller 20
consisting of broadband noise and a -3 dB tone at 100 Hz. The top trace is
the power spectrum of the channel input. In this case there is no
additional additive noise, so the middle trace is the channel output, and
the lower trace is the canceller output. Note that the canceller 20 is
stable and achieves in excess of 40 dB of suppression.
For example, suppose it is desired to operate the canceller 20 in the band
from 170 to 400 Hz. Without delay, the LMS canceller is unstable. However,
from FIG. 3, there exists a range of delays which adequately equalize the
phase response for in-band stability. It is easy to show that stability is
achieved with delay values ranging from 0.6 to 1.7 milliseconds. This
range of values achieves stability with a broad range of delays. For a
sampling frequency of 10 k Hz (used in the computer model), the delays
correspond to from 6 to 17 sample delays. Insertion of the 13 sample delay
has provided sufficient bending and leveling of the phase response of the
speaker-microphone transfer function to extend the stability region to the
band 170 Hz to 600 Hz.
Simulations of the filter using random inputs are also presented to support
these analytical performance predictions. In the simulations, a 6-tap low
pass FIR filter represented the acoustic channel through which the signal
passed, modelling simple multipath propagation. White Gaussian noise was
added to the output of this filter to represent the ambient background.
Many simulation cases have been made using this model, encompassing
ensembles of the noise processes as well as the full range of added delay
values. Some typical sample cases are presented below with reference to
FIGS. 6-10. The signals were modelled as a single frequency carrier,
modulated with narrow-band random processes of different bandwidths and
modulations. The ambient noise levels were set at -30 dB below the signal
levels. The solid lines in these figures represent the channel output
power while the dashed lines represent the cancelled output power.
The bandwidth of the input narrowband process and center frequency was set
at 5 Hz and 200 Hz, respectively, in the first sample run shown in FIG. 6.
A 64 tap FIR filter configuration is used with adaptation constant of
10.sup.-3. Rapid convergence of the error waveform to the noise floor was
achieved in less than 0.1 second. The parameters of the second sample run
shown in FIG. 7 were identical to the first run except the center
frequency of the narrowband process was modulated linearly in time at a
rate of 50 Hz/sec. Almost identical convergence characteristics were
achieved in the second run.
The input signal waveform parameters in the next case shown in FIG. 8 was
as in the first two cases except the bandwidth of the narrowband process
is increased to 20 Hz. The adaptation constant and filter tap size were
changed to 4.times.10.sup.-4 and 128, respectively, for better
cancellation performance. This also demonstrates successful adaptive
removal of the unwanted signals down to the level of the background noise.
However, due to the broader bandwidths of the signals to be cancelled, the
adaptive filter converged more slowly than in the first two runs.
Nevertheless, significant (20 dB or more) cancellation was achieved in
less than one second for both cases.
Finally, in the last sample run shown in FIG. 9, the signal parameters are
the same as in the first run except the filter is updated with only 5
units of delay. Instead of dropping to the -30 dB noise floor as in the
previous cases, the canceller output power grows rapidly without bound,
indicating that the LMS algorithm becomes unstable with a 5 sample delay
as theory predicts. The adaptation constants and adaptive filter tap sizes
were varied for this delay value. All variations have resulted in
algorithm instability. Thus the simulations have supported the analytical
prediction that the canceller is unstable for delays less than 5 samples,
and that there is a large range of delays (from 6 to 17) for which the
algorithm is stable.
Thus there has been described new and improved active adaptive noise
cancellers that do not require a training mode. It is to be understood
that the above-described embodiment is merely illustrative of some of the
many specific embodiments which represent applications of the principles
of the present invention. Clearly, numerous and other arrangements can be
readily devised by those skilled in the art without departing from the
scope of the invention.
Top